Quantization format

Q6_K_L

Q6_K_L has a nominal rate of bits per weight. We have no measured files carrying this label yet.

From the file· 526 files measured
Nominal bpw
from the block layout
Measured average
6.811
526 files
Range
4.606–9.239
varies by architecture
File sizes
0.11 GiB+
up to 325.38 GiB

Real files

one per model, most downloaded first
ModelSizeEffective bpwvs nominal
Qwen3.6-35B-A3B29.05 GiB6.941
gemma-4-26B-A4B-it21.46 GiB6.944
gemma-4-12B-it9.76 GiB7.012
Qwen3.5-4B3.69 GiB6.797
gemma-4-31B-it25.21 GiB6.924
Llama-3.2-1B-Instruct1.01 GiB7.026
gemma-4-E2B-it4.25 GiB7.128
Qwen3-8B6.54 GiB6.864
Qwen3.5-0.8B0.70 GiB6.897
Llama-3.1-8B-Instruct6.38 GiB6.825
Ornith-1.0-35B28.22 GiB6.994
Llama-3.2-3B-Instruct2.55 GiB6.821
ThinkingCap-Qwen3.6-27B22.43 GiB7.042
Qwythos-9B-v27.76 GiB6.903
Qwen3.5-122B-A10B101.31 GiB6.957
Qwen3-Coder-Next61.41 GiB6.621
Qwen2.5-32B-Instruct25.39 GiB6.657
Qwen3.5-35B-A3B29.05 GiB6.941
Qwen2.5-Coder-7B-Instruct6.07 GiB6.847
Qwen2.5-7B-Instruct6.07 GiB6.847
Qwen3-0.6B0.65 GiB7.430
Qwen3-VL-2B-Instruct1.39 GiB5.614
Jan-v3-4B-base-instruct3.55 GiB6.916
Qwen2.5-1.5B-Instruct1.24 GiB6.889
Qwen2.5-VL-7B-Instruct6.07 GiB6.288
From the filewhat these mean

Other quantizations